Method and equipment for positioning and tracking of variable-scale moving targets by integrating multiple features

Abstract

The present invention discloses a variable-scale moving target positioning and tracking method and equipment that integrates multiple features, relating to the field of image processing technology. The method comprises the following steps: obtaining feature points of a workpiece image and establishing a feature point matching relationship between adjacent workpiece image frames; determining a preliminary workpiece contour region and classifying the workpiece based on the preliminary workpiece contour region to obtain a final workpiece contour region; determining contour parameters; and dynamically positioning and tracking the workpiece by combining the workpiece's pose vector with its three-dimensional point cloud. Through a multi-scale feature fusion algorithm, the present invention can extract richer and more detailed feature points from the workpiece image, enhancing the ability to capture subtle features of the workpiece and facilitating the establishment of more accurate feature point matching relationships between adjacent image frames, thereby improving overall positioning accuracy.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and equipment for positioning and tracking a variable-scale moving target by integrating multiple features. Background Art

[0002] In complex industrial environments, the continuous movement of identified workpieces on the assembly line poses a challenge to real-time tracking and positioning. In order to effectively achieve this goal, it is particularly important to study the feature extraction technology of variable-scale moving targets that integrates multiple features. This technology needs to be able to adapt to the ever-changing environment and the dynamics of the target to ensure accurate identification and tracking of the workpiece. The dynamic tracking system should be able to automatically adjust its parameters to adapt to targets of different scales, thereby optimizing the tracking process. This includes using deep learning networks to automatically identify the most effective features and adjusting the network structure in real time to adapt to changes in target size. Ultimately, this system will be able to provide continuous and accurate workpiece position data, support automated operation and monitoring on the assembly line, and greatly improve production efficiency and safety.

[0003] In the existing technology, it is not convenient to extract rich and detailed feature points from the workpiece image, which reduces the ability to capture subtle features of the workpiece, and it is not convenient to establish a more accurate feature point matching relationship between adjacent image frames, thereby reducing the overall positioning accuracy. It is not convenient to perform preliminary detection of the workpiece contour, nor is it convenient to optimize the accuracy of contour parameters, and the ability to handle abnormal values is reduced. At the same time, it is not convenient to estimate the pose vector of the workpiece, and it is not convenient to use the pose vector and three-dimensional point cloud data to achieve dynamic tracking. It cannot effectively cope with the high-speed movement and position changes of the workpiece on the production line, and cannot ensure the continuity and real-time performance of tracking.

[0004] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0005] In response to the deficiencies in the prior art, the present invention proposes a variable-scale moving target positioning and tracking method and equipment that integrates multiple features, which solves the problems raised in the above background technology, such as the inconvenience in extracting rich and detailed feature points from the workpiece image, reduced ability to capture subtle features of the workpiece, and inconvenience in establishing a more accurate feature point matching relationship between adjacent image frames, thereby reducing the overall positioning accuracy, making it inconvenient to perform preliminary detection of the workpiece contour, and also making it inconvenient to optimize the accuracy of contour parameters, and reducing the ability to handle outliers. At the same time, it is inconvenient to estimate the pose vector of the workpiece, and it is inconvenient to use the pose vector and three-dimensional point cloud data to achieve dynamic tracking, and it cannot effectively cope with the high-speed movement and position changes of the workpiece on the production line, and cannot ensure the continuity and real-time performance of tracking.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] According to one aspect of the present invention, a method for positioning and tracking a variable-scale moving target by fusing multiple features is provided. The method for positioning and tracking a variable-scale moving target comprises the following steps:

[0008] S1. Acquire an image of the workpiece to be marked, extract features from the workpiece image using a multi-scale feature fusion algorithm, obtain feature points of the workpiece image, and establish a feature point matching relationship between adjacent workpiece image frames;

[0009] S2. Perform preliminary detection on the contour of the workpiece image using Hough transform, determine a preliminary workpiece contour area, and classify the workpiece categories based on the preliminary workpiece contour area to obtain a final workpiece contour area;

[0010] S3, using the double point removal least squares method to regress the feature points in the final workpiece contour area to determine the contour parameters;

[0011] S4. Obtain the three-dimensional point cloud and workpiece image features of the workpiece, use a deep convolutional neural network to estimate the workpiece pose vector, and combine the workpiece pose vector with the three-dimensional point cloud to perform dynamic positioning and tracking of the workpiece;

[0012] Acquiring an image of the workpiece to be marked, extracting features from the workpiece image using a multi-scale feature fusion algorithm, obtaining feature points of the workpiece image, and establishing a feature point matching relationship between adjacent workpiece image frames includes the following steps:

[0013] S11, extracting feature points from the acquired workpiece image using a multi-scale feature fusion algorithm;

[0014] S12, extracting stable feature points from the workpiece image, and calculating the descriptor of each feature point to characterize the image information around the feature point;

[0015] S13, matching feature points in two adjacent workpiece images, and realizing feature point matching relationship by calculating the distance and similarity between the two descriptors.

[0016] Furthermore, extracting feature points from the acquired workpiece image using a multi-scale feature fusion algorithm includes the following steps:

[0017] S111, performing multi-scale transformation on the workpiece image by using a Gaussian difference pyramid, and generating workpiece images of different scale spaces;

[0018] S112, detecting local extreme points as candidate feature points by comparing adjacent pixels in each scale-space workpiece image;

[0019] S113, calculating the gradient and direction of the surrounding pixels for each candidate feature point, generating a gradient direction histogram, and determining the main direction of the feature point based on the peak value of the histogram;

[0020] S114. Construct a local area with each feature point as the center, calculate the gradient direction and amplitude in the local area, and combine them into a descriptor vector as the descriptor of the feature point to perform feature matching.

[0021] Furthermore, the contour of the workpiece image is preliminarily detected using Hough transform to determine the preliminary workpiece contour area, and the workpiece categories are divided based on the preliminary workpiece contour area to obtain the final workpiece contour area, which includes the following steps:

[0022] S21, converting the workpiece color image into a grayscale image, and applying Gaussian filtering to remove image noise;

[0023] S22, extracting edges in the grayscale image using an edge detection algorithm;

[0024] S23, detecting the shape of the workpiece according to the need and setting the parameter space;

[0025] S24, increasing the count of all parameters corresponding to each edge point on the grayscale image in the parameter space to form a cumulative array, and searching for a local maximum in the cumulative array;

[0026] S25, obtaining a peak value from the accumulated array as a parameter of the detected workpiece shape, and using the parameter of the workpiece shape to draw a contour on the original workpiece image as a preliminary workpiece contour area;

[0027] S26. Based on the preliminary workpiece contour area, the workpiece category is divided using a division algorithm to obtain a final workpiece contour area.

[0028] Furthermore, based on the preliminary workpiece contour area, the workpiece category is divided using a division algorithm to obtain the final workpiece contour area, including the following steps:

[0029] S261, setting the scaling factor and cross factor of the partitioning algorithm, and determining the maximum number of iterations and the number of chaotic searches;

[0030] S262, classifying the workpiece into categories according to the characteristics of the workpiece image and initializing candidate solutions for the workpiece contour;

[0031] S263, randomly selecting a workpiece contour candidate solution from the workpiece contour candidate solutions according to each workpiece category to generate a new variant workpiece contour candidate solution;

[0032] S264, fusing different workpiece contour candidate solutions through cross factors to generate a new workpiece contour candidate solution;

[0033] S265, comparing the newly generated candidate workpiece contour solution with the current optimal workpiece contour solution, and retaining the optimal workpiece contour solution;

[0034] S266, repeating steps S263 to S265 until a predetermined number of iterations is reached;

[0035] S267. Perform a chaotic search around the optimal workpiece contour solution of each category, and select the workpiece contour with the highest fitness from the chaotic search as the final workpiece contour region.

[0036] Furthermore, the formula for generating new candidate solutions for variant workpiece contours is:

[0037] ;

[0038] Where, A i ( m +1) indicates the next iteration m +1 at the time i Candidate solutions for variant workpiece contours;

[0039] G Expressed as a scaling factor;

[0040] C r1 ( m )、 C r2 ( m )and C r2 ( m ) are respectively represented as m A candidate solution of the workpiece contour randomly selected from the candidate solutions of the workpiece contour at the time;

[0041] r 1. r 2 and r 3 are random variables.

[0042] Furthermore, the feature points in the final workpiece contour area are regressed using the double-point removal least squares method to determine the contour parameters, which includes the following steps:

[0043] S31, obtaining a feature point data set within the final workpiece contour area;

[0044] S32, using the least squares method to perform preliminary linear regression analysis on all feature points to obtain initial contour parameter estimates;

[0045] S33, removing two feature points in the feature point data set one by one, and re-performing least squares regression using the remaining feature points;

[0046] S34, calculating the difference between the contour parameters obtained by least squares regression after removing two feature points each time and the initial contour parameter estimate, and recording the difference value;

[0047] S35, repeating step S33 and step S34 until the preset number of iterations is reached;

[0048] S36. Analyze the difference values of all records and select a set of regression results with the smallest difference values as the estimate of the final profile parameters.

[0049] Furthermore, analyzing the difference values of all records and selecting a set of regression results with the smallest difference values as the final estimation of the profile parameters includes the following steps:

[0050] S361. Obtain all historical regression result datasets and initialize the regression result datasets and related constraints;

[0051] S362, accepting new regression result data update instructions and selecting a data processing mode according to analysis requirements;

[0052] S363. If the priority of minimizing the difference value is selected, the constraint boundary of the data set is calculated using a specific formula. If the priority of other is selected, the original data set boundary remains unchanged.

[0053] S364. Calculate the deviation of each set of regression result data and the deviation of the entire regression result data set, and select data points that need to be adjusted first;

[0054] S365. Perform regression analysis on the selected data points using an optimization algorithm, and record the difference value of each optimization;

[0055] S366. Analyze the difference values of all records, compare the calculated difference values, and select a set of regression results with the smallest difference value as the final profile parameters.

[0056] Furthermore, performing regression analysis optimization on the selected data points using an optimization algorithm and recording the difference value of each optimization includes the following steps:

[0057] S3651. Randomly initialize several sets of regression model parameters, each set of parameters representing a potential regression model solution;

[0058] S3652. Calculate the fitness of each set of regression model parameters, and select the regression model with the smallest difference value as the core regression model;

[0059] S3653. Adjust the non-core regression model parameter set, adopt different adjustment strategies for the core regression model parameter set, and determine the number of new parameters to be generated based on their respective difference values;

[0060] S3654. Generate a new parameter set based on the current regression model parameters and perform boundary detection;

[0061] S3655, applying Gaussian mutation to selected regression model parameters to generate additional parameter sets;

[0062] S3656. From all current regression model parameters, select a set of parameters with the smallest difference value to continue iteration, and record the best difference value;

[0063] S3657. Check whether the predetermined number of iterations has been reached. If so, stop the iteration and record the minimum difference value as the regression result.

[0064] Furthermore, obtaining the three-dimensional point cloud and image features of the workpiece, estimating the workpiece pose vector using a deep convolutional neural network, and dynamically positioning and tracking the workpiece by combining the workpiece pose vector and the three-dimensional point cloud of the workpiece include the following steps:

[0065] S41. Use deep residual network to extract features from workpiece images and design pose estimation network;

[0066] S42, transforming the output of the feature extraction part through a fully connected layer to regress the pose vector;

[0067] S43, dividing the output of the pose regression part into a position vector and a pose vector, outputting the estimated poses of the multiple classes of objects respectively;

[0068] S44, collect 3D point cloud and image data of the workpiece in real time in a dynamic environment, and use the trained deep residual network to continuously update the workpiece pose estimation;

[0069] S45. Combine the real-time updated pose vector and 3D point cloud to implement accurate dynamic positioning and tracking of workpieces.

[0070] According to another aspect of the present invention, a multi-feature integrated variable-scale moving target positioning and tracking device is provided. The multi-feature integrated variable-scale moving target positioning and tracking device includes:

[0071] The image processing module is used to obtain the image of the workpiece to be marked, extract features from the workpiece image using a multi-scale feature fusion algorithm, obtain feature points of the workpiece image, and establish a feature point matching relationship between adjacent workpiece image frames;

[0072] The workpiece detection module is used to perform preliminary detection on the contour of the workpiece image using Hough transform, determine the preliminary workpiece contour area, and classify the workpiece categories based on the preliminary workpiece contour area to obtain the final workpiece contour area;

[0073] The data analysis and regression model module is used to regress the feature points in the final workpiece contour area using the double-point removal least squares method to determine the contour parameters;

[0074] The target positioning and tracking module is used to obtain the workpiece's 3D point cloud and workpiece image features, estimate the workpiece's pose vector using a deep convolutional neural network, and dynamically position and track the workpiece by combining the workpiece's pose vector with the workpiece's 3D point cloud;

[0075] Among them, the image processing module is connected through the workpiece detection module and the data analysis and regression model module, and the data analysis and regression model module is connected with the target positioning and tracking module.

[0076] The beneficial effects of the present invention are:

[0077] 1. Through the multi-scale feature fusion algorithm, the present invention can extract richer and more detailed feature points from the workpiece image, enhance the ability to capture subtle features of the workpiece, and help establish a more accurate feature point matching relationship between adjacent image frames, thereby improving the overall positioning accuracy. The application of Hough transform in contour detection can provide a stable preliminary contour area for workpiece classification and tracking, which helps to maintain efficient operation under conditions of visual noise or partial occlusion. The double-point removal least squares rule further optimizes the accuracy of contour parameters and enhances the system's ability to handle outliers by eliminating abnormal data points. The deep convolutional neural network is combined to estimate the workpiece's pose vector, and the pose vector and three-dimensional point cloud data are used to achieve dynamic tracking, which can effectively cope with the high-speed movement and position changes of the workpiece on the production line, ensuring the continuity and real-time tracking.

[0078] 2. The present invention uses a multi-scale feature fusion algorithm to capture image details and structures at different scale levels. This method is extremely effective for detecting feature points that remain stable under scale transformations. This capability is particularly suitable for scenarios where the size, shape, or orientation of the workpiece changes rapidly. The descriptor constructed for each feature point by calculating the gradient and direction of its surrounding pixels not only provides the location information of the point, but also accurately describes its local image features. The distance and similarity between descriptors are used to match feature points, making the matching process more robust, especially when the image is slightly distorted or occluded due to the movement of the workpiece. This can effectively connect the same workpiece in consecutive image frames, providing stable input for dynamic tracking.

[0079] 3. This invention enhances the accurate recognition of workpiece contours by utilizing the Hough transform in combination with multi-step edge detection and shape parameter extraction, especially when image quality is affected by noise. Grayscale conversion and Gaussian filtering help reduce processing errors and ensure the accuracy of edge detection. The partitioning algorithm used in the classification process further optimizes the identification of workpiece categories, ensuring that each type of workpiece is correctly classified. Utilizing chaotic search technology, the workpiece contour is optimized and adjusted after the classification is determined. This not only improves tracking accuracy but also increases adaptability to dynamic changes. By integrating different solutions and adjusting algorithm parameters such as scaling factors and crossover factors, it can flexibly respond to the diversity and complexity of workpiece visual presentation, thereby maintaining high performance in a constantly changing industrial environment.

[0080] 4. The present invention uses the least squares method for preliminary regression to provide a robust baseline. By removing potential abnormal feature points pair by pair and recalculating the regression, the influence of noise and outliers can be significantly reduced, and the accuracy and robustness of parameter estimation can be improved. By continuously analyzing the difference values and selecting the regression result with the smallest difference value, the model is guaranteed to perform optimally when faced with variable and complex data. This method is particularly suitable for dynamically changing industrial environments, ensuring that the estimation of profile parameters can adapt to actual physical changes. By meticulously managing each data point and iterative process, it can more effectively process large amounts of data, reduce the waste of computing resources, and thus optimize operational efficiency and output quality.

[0081] 5. This invention uses a deep residual network to extract high-quality features from workpiece images and a pose estimation network to accurately calculate pose vectors, thereby achieving efficient dynamic tracking. In dynamic environments, it can collect and process 3D point cloud and image data in real time, continuously updating the workpiece pose estimate. This not only improves the system's adaptability to environmental changes, but also ensures tracking continuity and accuracy under various operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0083] Figure 1 is a flowchart of a method for positioning and tracking a variable-scale moving target by fusing multiple features according to an embodiment of the present invention;

[0084] Figure 2 It is a principle block diagram of a variable-scale moving target positioning and tracking device integrating multiple features according to an embodiment of the present invention.

[0085] In the picture:

[0086] 1. Image processing module; 2. Workpiece detection module; 3. Data analysis and regression model module; 4. Target positioning and tracking module. DETAILED DESCRIPTION

[0087] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0088] In the description of the present invention, unless otherwise specified, "plurality" means two or more. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0089] According to an embodiment of the present invention, a method and equipment for positioning and tracking a variable-scale moving target by fusing multiple features are provided.

[0090] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, a method for positioning and tracking a variable-scale moving target by integrating multiple features includes the following steps:

[0091] S1. Acquire an image of the workpiece to be marked, extract features from the workpiece image using a multi-scale feature fusion algorithm, obtain feature points of the workpiece image, and establish a feature point matching relationship between adjacent workpiece image frames;

[0092] S2. Perform preliminary detection on the contour of the workpiece image using Hough transform, determine a preliminary workpiece contour area, and classify the workpiece categories based on the preliminary workpiece contour area to obtain a final workpiece contour area;

[0093] S3, using the double point removal least squares method to regress the feature points in the final workpiece contour area to determine the contour parameters;

[0094] S4. Obtain the three-dimensional point cloud and workpiece image features of the workpiece, use a deep convolutional neural network to estimate the workpiece pose vector, and combine the workpiece pose vector with the three-dimensional point cloud to perform dynamic positioning and tracking of the workpiece;

[0095] It should be explained that the workpiece is continuously photographed by a camera, capturing images. Each workpiece image is processed using multi-scale feature fusion algorithms, such as the scale-invariant feature transformation algorithm, to extract key feature points. This algorithm detects local extreme points in different scale spaces and calculates their orientation and gradient, providing a scale- and rotation-invariant feature description. This algorithm is suitable for maintaining feature stability at multiple scales. The edges and contours of each workpiece image are detected using the Hough transform. After the preliminary contour is determined, a shape-based classification algorithm (such as a support vector machine or chaotic differential evolution algorithm) is used to classify the workpiece to determine its type. After preliminary classification, feature points within the contour region of workpieces belonging to the same category are subjected to regression analysis using the double-point removal least squares method to determine more accurate contour parameters. On the assembly line, a 3D scanner acquires 3D point cloud data of the workpiece in real time. Combined with the pose vector estimated by a deep convolutional neural network, the position and orientation of the workpiece on the assembly line can be tracked in real time.

[0096] Acquiring an image of the workpiece to be marked, extracting features from the workpiece image using a multi-scale feature fusion algorithm, obtaining feature points of the workpiece image, and establishing a feature point matching relationship between adjacent workpiece image frames includes the following steps:

[0097] S11, extracting feature points from the acquired workpiece image using a multi-scale feature fusion algorithm;

[0098] S12, extracting stable feature points from the workpiece image, and calculating the descriptor of each feature point to characterize the image information around the feature point;

[0099] S13, matching feature points in two adjacent workpiece images, and realizing feature point matching relationship by calculating the distance and similarity between the two descriptors.

[0100] For clarification, assume that a factory captures multiple consecutive images of a chip (workpiece) using a camera. Because chips may have complex textures or subtle identifying marks, a multi-scale feature fusion algorithm (such as the scale-invariant feature transform algorithm, the accelerated robust feature algorithm, or the ORB algorithm) is used to extract feature points from the chip images. These feature points can be edges, marks, or other prominent regions on the chip. The purpose of a multi-scale feature fusion algorithm is to extract stable feature points from different image scales. For example, the scale-invariant feature transform algorithm decomposes the image into multiple scale spaces using a Gaussian difference pyramid and searches for local extreme points within these spaces as feature points. This method is robust to rotation, scaling, and illumination variations, making it well-suited for chip image processing in dynamic industrial environments. For each feature point extracted from the chip image, a descriptor is calculated. This descriptor characterizes the image structure and information surrounding the feature point. The descriptor ensures uniqueness for the feature point, ensuring that it remains identifiable in other images even if the chip image is slightly deformed or captured from a different angle. A descriptor is a vector encoding that represents the local information of a feature point. For example, the scale-invariant feature transform algorithm descriptor encodes the local features of a feature point by calculating the gradient direction histogram of the area surrounding the feature point. These descriptors are crucial for matching feature points and eliminating noise. Suppose that in two consecutive image frames, the first and second frames contain the same chip. Feature points are matched by comparing the feature point descriptors in these frames. Typically, the matching process determines their similarity by calculating the Euclidean distance between two descriptors. The higher the similarity, the more likely the two feature points are from the same location. By using the matching relationship between feature points, the system can track the movement of chips on the assembly line. Feature point matching is achieved by calculating the distance between descriptors. The most common method is to use the Euclidean distance or other similarity metrics. For example, the nearest neighbor algorithm can quickly match two feature points. To eliminate false matches, a "bidirectional matching" strategy is often used, ensuring that feature points A and B are the best match for each other.

[0101] Preferably, extracting feature points from the acquired workpiece image using a multi-scale feature fusion algorithm includes the following steps:

[0102] S111, performing multi-scale transformation on the workpiece image by using a Gaussian difference pyramid, and generating workpiece images of different scale spaces;

[0103] Specifically, performing multi-scale transformation on the workpiece image by using a Gaussian difference pyramid and generating workpiece images of different scale spaces includes the following steps:

[0104] S1111, performing Gaussian filtering on the workpiece image to obtain Gaussian blurred images of different scales;

[0105] S1112, performing a downsampling operation on each Gaussian blurred image to obtain low-resolution images of different scales;

[0106] S1113 , performing a subtraction operation on each adjacent low-resolution image to obtain Gaussian difference images of different scales as spatial artifact images of different scales.

[0107] It's important to explain that the camera captures the original image of the workpiece. To process workpiece images of varying scales, each image is first Gaussian filtered. The Gaussian filtering process smoothes the image and reduces noise through convolution. Different standard deviations are set to generate images with varying degrees of blur. The larger the standard deviation, the greater the blur. Each Gaussian-filtered image is then downsampled, reducing its resolution. For example, every other pixel in the original image is sampled to generate a smaller image. This process rapidly generates multi-scale, low-resolution images representing the workpiece from different perspectives. Downsampling allows for the generation of images of varying resolutions, adapting to the varying visual effects of varying distances or sizes of the workpiece in real-world scenarios. A pixel-by-pixel subtraction operation is performed on each adjacent low-resolution image to generate a Gaussian difference image. This step emphasizes edges and feature areas in the workpiece image through the difference operation, eliminating irrelevant information from the image.

[0108] S112, detecting local extreme points as candidate feature points by comparing adjacent pixels in each scale-space workpiece image;

[0109] S113, calculating the gradient and direction of the surrounding pixels for each candidate feature point, generating a gradient direction histogram, and determining the main direction of the feature point based on the peak value of the histogram;

[0110] S114. Construct a local area with each feature point as the center, calculate the gradient direction and amplitude in the local area, and combine them into a descriptor vector as the descriptor of the feature point to perform feature matching.

[0111] It's important to note that the camera captures multiple images of the workpiece, which may have different scales due to the distance or angle of the workpiece. Using a Gaussian difference pyramid, the workpiece image is Gaussian blurred multiple times, generating an image pyramid with varying blur levels. The blur level of these images gradually increases, starting with the original image. This allows the system to detect stable feature points at different scales. At each scale, each pixel is compared with its neighbors to identify local extreme points. A local extreme point is one where the value is larger or smaller than that of the surrounding pixels. These local extreme points are considered candidate feature points because they exhibit good stability across scales. After detecting a candidate feature point, the gradient and direction of each pixel in the neighborhood of the feature point are calculated to form a gradient direction histogram. This histogram summarizes the gradient strength in each direction within the region. To ensure rotational invariance, the system selects the peak in the gradient direction histogram as the dominant direction of the feature point. For each feature point, a local region (typically 16×16 pixels) is constructed with the peak value as the center. Within this local region, the gradient direction and magnitude of each pixel are calculated and combined into a feature vector, called a descriptor. Each descriptor represents the local structural information of that feature point. In the subsequent image matching process, descriptors are compared to determine whether the feature points in the two images are identical.

[0112] Preferably, performing preliminary detection on the contour of the workpiece image using Hough transform, determining a preliminary workpiece contour area, and classifying the workpiece categories based on the preliminary workpiece contour area to obtain the final workpiece contour area includes the following steps:

[0113] S21, converting the workpiece color image into a grayscale image, and applying Gaussian filtering to remove image noise;

[0114] S22, extracting edges in the grayscale image using an edge detection algorithm;

[0115] S23, detecting the shape of the workpiece according to the need and setting the parameter space;

[0116] S24, increasing the count of all parameters corresponding to each edge point on the grayscale image in the parameter space to form a cumulative array, and searching for a local maximum in the cumulative array;

[0117] S25, obtaining a peak value from the accumulated array as a parameter of the detected workpiece shape, and using the parameter of the workpiece shape to draw a contour on the original workpiece image as a preliminary workpiece contour area;

[0118] S26. Based on the preliminary workpiece contour area, the workpiece category is divided using a division algorithm to obtain a final workpiece contour area.

[0119] It's important to explain that after the camera captures a color image of the workpiece, it first converts it to a grayscale image. Grayscale images remove color information while retaining brightness information, making edges and shapes clearer. The system then applies a Gaussian filter to the grayscale image to remove high-frequency noise, such as surface texture and lighting variations, to ensure the accuracy of subsequent contour detection. Classic edge detection algorithms, such as Canny edge detection, are then used to extract edges. Canny edge detection calculates image gradients to identify areas with significant brightness changes, providing edge information for the workpiece. To detect circular workpieces, a Hough transform parameter space is established to define the circle's parameters (such as its center and radius). The Hough transform converts between image space and parameter space, recording the parameter values of each possible circle. At each edge point, the Hough transform algorithm increments a count for each possible circle parameter value within the parameter space. As all edge points vote, a peak forms in the cumulative array in the parameter space, indicating the presence of a circle. The system then finds a local maximum within the cumulative array, and the parameters corresponding to this maximum are the center and radius of the circular workpiece. Based on these parameters, a circular outline is drawn on the original image, marking the preliminary contour area of the workpiece. After obtaining the preliminary contour of the workpiece, a geometric feature-based classification algorithm (such as K-means clustering, support vector machine, chaotic differential evolution algorithm, etc.) is used to classify the workpiece based on its size and shape parameters. For example, a larger circle may be a bearing, while a smaller circle may be a washer. Ultimately, the workpiece category is determined based on the classification results, and the contour area is refined.

[0120] Preferably, based on the preliminary workpiece contour area, using a division algorithm to divide the workpiece categories to obtain the final workpiece contour area includes the following steps:

[0121] S261, setting the scaling factor and cross factor of the partitioning algorithm, and determining the maximum number of iterations and the number of chaotic searches;

[0122] S262, classifying the workpiece into categories according to the characteristics of the workpiece image and initializing candidate solutions for the workpiece contour;

[0123] S263, randomly selecting a workpiece contour candidate solution from the workpiece contour candidate solutions according to each workpiece category to generate a new variant workpiece contour candidate solution;

[0124] S264, fusing different workpiece contour candidate solutions through cross factors to generate a new workpiece contour candidate solution;

[0125] S265, comparing the newly generated candidate workpiece contour solution with the current optimal workpiece contour solution, and retaining the optimal workpiece contour solution;

[0126] S266, repeating steps S263 to S265 until a predetermined number of iterations is reached;

[0127] S267. Perform a chaotic search around the optimal workpiece contour solution of each category, and select the workpiece contour with the highest fitness from the chaotic search as the final workpiece contour region.

[0128] Specifically, the partitioning algorithm is based on the improved chaotic differential evolution algorithm. In order to accelerate the global convergence speed, the Logistic chaotic mapping is introduced. The chaotic differential evolution algorithm is realized by using the ergodicity of chaotic motion to perform chaotic search in the solution space centered on the best individual.

[0129] It is important to explain the initialization parameters for the partitioning algorithm. The scaling factor controls the degree of variation in candidate solutions, and the crossover factor controls the degree of fusion between different candidate solutions. Data example: scaling factor = 0.5, crossover factor = 0.9, maximum number of iterations = 100, and number of chaotic searches = 50. Based on the preliminary classification of the workpiece in the image, candidate contour solutions are initialized. Data example: Workpiece images are divided into three categories: circular, square, and irregular. For each category, a preliminary contour solution is extracted. For example, the initial contour for a circular workpiece can be an estimate of the radius, for a square workpiece it can be an estimate of the side length, and for an irregular shape, a new contour solution is generated through mutation based on the edge detection algorithm. Data example: For a circular workpiece, initial candidate solutions with radii of 20 mm, 25 mm, and 30 mm are selected. Mutation generates a new radius solution = 25 + scaling factor × (30 - 20), resulting in a new radius = 25 + 0.5 × 10 = 30. Similarly, for square workpieces, side lengths can be mutated to generate new candidate solutions. A crossover operation can be used to combine different categories of contour solutions to generate a new candidate solution. For example, using a crossover factor of 0.9, different parameters are fused from the existing contour solutions with a 90% probability. For circular workpieces, different radii can be fused, while for square workpieces, contour solutions with different side lengths may be fused. The fitness of the newly generated contour solutions is evaluated, and the solution with the highest fitness is retained as the current optimal solution. For example, the fitness function can be based on the degree of match between the contour solution and the actual workpiece image contour; the higher the match, the higher the fitness. Assuming the fitness of the current optimal solution is 0.85 and the fitness of the new solution is 0.88, the new solution is retained as the optimal solution, and the contour solution is gradually optimized through iteration. For example, the mutation and crossover operations are repeated, each time generating a new contour solution and comparing it with the current optimal solution. After 100 iterations, the fitness reaches 0.92. A chaotic search algorithm is used to perform local search near the optimal solution to further optimize the contour solution. Data example: a local search is performed near the optimal solution radius = 30. Contour solutions with radii of 29mm, 30mm, and 31mm are generated through chaotic search. After evaluating the fitness, it is found that 30.5mm has the highest fitness, and it is finally selected as the final workpiece contour solution.

[0130] Preferably, the formula for generating a new candidate solution for the variant workpiece contour is:

[0131] ;

[0132] Where, A i ( m +1) indicates the next iteration m +1 at the time i Candidate solutions for variant workpiece contours;

[0133] G Expressed as a scaling factor, G belongs to (0, 1);

[0134] C r1 ( m )、 C r2 ( m )and C r2 ( m ) are respectively represented as m A candidate solution of the workpiece contour randomly selected from the candidate solutions of the workpiece contour at the time;

[0135] r 1. r 2 and r 3 are random variables (which are random positive integers and satisfy r 1≠ r 2≠ r 3≠ i ).

[0136] Preferably, regressing the feature points in the final workpiece contour area using the double-point removal least squares method to determine the contour parameters includes the following steps:

[0137] S31, obtaining a feature point data set within the final workpiece contour area;

[0138] S32, using the least squares method to perform preliminary linear regression analysis on all feature points to obtain initial contour parameter estimates;

[0139] S33, removing two feature points in the feature point data set one by one, and re-performing least squares regression using the remaining feature points;

[0140] S34, calculating the difference between the contour parameters obtained by least squares regression after removing two feature points each time and the initial contour parameter estimate, and recording the difference value;

[0141] S35, repeating step S33 and step S34 until the preset number of iterations is reached;

[0142] S36. Analyze the difference values of all records and select a set of regression results with the smallest difference values as the estimate of the final profile parameters.

[0143] It should be noted that the gear's contour region has been identified and extracted through the previous contour detection and shape segmentation. All feature points within this contour region (such as edge points and key contour points) are combined into a dataset for subsequent contour fitting. All extracted feature points are then input into a least squares (Least Squares) model for linear regression. Assuming the gear's contour is approximately circular or elliptical, a preliminary regression is performed on the data using an appropriate curve model (such as circle or ellipse fitting) to obtain an initial estimate of the contour parameters. After the preliminary regression is complete, the system begins pairwise feature point removal. Each time, two feature points are removed from the feature point dataset, and the remaining feature points are re-entered into the least squares model for regression. The two removed feature points may be outliers or points that significantly deviate from the regression results. Each time two feature points are removed, a new regression result is obtained. The difference between the contour parameters of this regression result and the initial regression result is calculated and recorded. A smaller difference indicates that the removed feature point has a significant impact on the regression result, possibly due to noise or an outlier. The process of removing feature points, re-regressing, and calculating the difference is repeated for a preset number of iterations (e.g., 500). Through multiple iterations, the system gradually eliminates the effects of noise and outliers on the regression results, further optimizing the fitting curve. After all iterations are complete, the difference values obtained from each regression are compared, and the one with the smallest difference value is selected as the optimal regression result. The resulting profile parameters are considered the final, accurate estimate. Based on these parameters, the gear profile is drawn. The regression result with the smallest difference value represents the most accurate fit after removing noise and outliers. By selecting the regression result with the smallest difference value, the accuracy of the profile parameters is ensured.

[0144] Preferably, analyzing the difference values of all records and selecting a set of regression results with the smallest difference values as the estimation of the final profile parameters includes the following steps:

[0145] S361. Obtain all historical regression result datasets and initialize the regression result datasets and related constraints;

[0146] S362, accepting new regression result data update instructions and selecting a data processing mode according to analysis requirements;

[0147] S363. If the priority of minimizing the difference value is selected, the constraint boundary of the data set is calculated using a specific formula. If the priority of other is selected, the original data set boundary remains unchanged.

[0148] S364. Calculate the deviation of each set of regression result data and the deviation of the entire regression result data set, and select data points that need to be adjusted first;

[0149] S365. Perform regression analysis on the selected data points using an optimization algorithm, and record the difference value of each optimization;

[0150] S366. Analyze the difference values of all records, compare the calculated difference values, and select a set of regression results with the smallest difference value as the final profile parameters.

[0151] It is important to explain that a set of historical regression results is obtained from past contour fitting. This data includes parameter estimates for each set of regression curves during the fitting process, as well as deviations from the initial contour estimates. Next, a regression result dataset is initialized and relevant constraints are set, such as maximum allowable deviation or limits on specific contour parameters (such as roundness and smoothness). Regression analysis is run based on the latest contour feature point data, generating new regression result data. The system, under user or automated instructions, chooses how to process the new data: whether to prioritize minimizing discrepancies or to maintain the original dataset boundaries to preserve historical reference. If minimizing discrepancies is prioritized, a specific formula is used to recalculate the regression dataset boundaries, such as using Lagrange multipliers or other optimization algorithms, to constrain the regression results to within the specified boundaries. Otherwise, the original constraints are maintained, allowing the regression results to remain consistent with the historical data. The deviations of each regression result are then analyzed step by step, specifically the differences between each set of regression parameters and the overall result. By analyzing the impact of each data point on the overall regression, we can identify those points with the greatest impact on deviation and mark them as priority points for adjustment. We then re-run the regression analysis using optimization algorithms (such as gradient descent, Newton's method, and fireworks) for these prioritized data points, adjusting the weights or positions of the data points to reduce the overall regression deviation. The difference values after each optimization are recorded for subsequent comparison and analysis. All recorded difference values are analyzed and compared, and the set of regression results with the smallest difference values is selected as the final profile parameter estimate. These profile parameters are used to generate the precise profile of the body part, ensuring that the processing equipment operates according to the optimized profile.

[0152] Preferably, performing regression analysis optimization on the selected data points using an optimization algorithm and recording the difference value of each optimization includes the following steps:

[0153] S3651. Randomly initialize several sets of regression model parameters, each set of parameters representing a potential regression model solution;

[0154] S3652. Calculate the fitness of each set of regression model parameters, and select the regression model with the smallest difference value as the core regression model;

[0155] S3653. Adjust the non-core regression model parameter set, adopt different adjustment strategies for the core regression model parameter set, and determine the number of new parameters to be generated based on their respective difference values;

[0156] S3654. Generate a new parameter set based on the current regression model parameters and perform boundary detection;

[0157] S3655, applying Gaussian mutation to selected regression model parameters to generate additional parameter sets;

[0158] S3656. From all current regression model parameters, select a set of parameters with the smallest difference value to continue iteration, and record the best difference value;

[0159] S3657. Check whether the predetermined number of iterations has been reached. If so, stop the iteration and record the minimum difference value as the regression result.

[0160] Specifically, the optimization algorithm is an improved fireworks algorithm. In the basic fireworks algorithm, the number of explosion sparks and the size of the explosion radius generated by an individual are completely determined by the difference in fitness between that individual and other individuals. In the later stages of the algorithm's optimization process, when the population approaches the optimal value, the numerator of the explosion operator approaches zero, that is, the explosion radius approaches zero, making subsequent optimization processes essentially ineffective. This limits its local optimization capabilities. Therefore, the basic fireworks algorithm has been improved. Specifically, the optimal explosion radius of individuals in the fireworks population is dynamically determined, rather than simply based on fitness changes. This improved fireworks algorithm possesses greater local exploration capabilities than the basic fireworks algorithm. Because its optimization radius changes dynamically and gradually decreases, the improved algorithm avoids the problem of optimization failure. Its local optimization accuracy is greatly improved, and in load distribution, it can assign the optimal load to each unit as much as possible.

[0161] It is important to explain that multiple sets of regression model parameters are first randomly generated. These parameters may represent different blade profile shapes (e.g., varying curvatures, etc.). For example, 10 sets of initial parameters may be randomly generated, each representing a possible blade profile. For each set of regression model parameters, its fitness is calculated, typically by comparing the error between the regression model fit and the actual blade feature points. A higher fitness indicates a lower error. The regression model with the highest fitness (lowest variance) is selected as the core regression model. All non-core regression model parameter sets are then adjusted, for example, through simulated annealing, genetic algorithm crossover, and mutation to generate new parameters. Different strategies are employed for the core regression model parameters, such as remaining unchanged or undergoing minor adjustments. Based on the variance, a new number of parameters is generated to ensure further exploration of the optimal parameter set. Based on the adjusted parameters, new parameter sets are generated and checked to ensure they fall within pre-set physical and engineering constraints (e.g., the blade curvature must not exceed the material limit). If a parameter exceeds the bounds, it is recalibrated to a reasonable range. Gaussian mutation is applied to the currently selected regression model parameters, generating new parameter sets by introducing small random variations (Gaussian noise) around the parameters. These new parameter sets further enrich the diversity of model solutions. All current parameter sets are analyzed, and the set with the smallest difference value is selected as the starting point for a new iteration. The best difference value from this iteration is recorded. This set of parameters represents the fit result that most closely matches the actual blade profile. The current number of iterations is checked to see if it has reached a predetermined limit (for example, 100 iterations). If so, the iterations are terminated, and the recorded minimum difference value and its corresponding regression model parameters are used as the final regression result to determine the blade profile parameters.

[0162] Preferably, obtaining the three-dimensional point cloud and workpiece image features of the workpiece, estimating the workpiece pose vector using a deep convolutional neural network, and dynamically positioning and tracking the workpiece by combining the workpiece pose vector and the three-dimensional point cloud of the workpiece includes the following steps:

[0163] S41. Use deep residual network to extract features from workpiece images and design pose estimation network;

[0164] S42, transforming the output of the feature extraction part through a fully connected layer to regress the pose vector;

[0165] S43, dividing the output of the pose regression part into a position vector and a pose vector, outputting the estimated poses of the multiple classes of objects respectively;

[0166] S44, collect 3D point cloud and image data of the workpiece in real time in a dynamic environment, and use the trained deep residual network to continuously update the workpiece pose estimation;

[0167] S45. Combine the real-time updated pose vector and 3D point cloud to implement accurate dynamic positioning and tracking of workpieces.

[0168] It should be noted that images of the workpiece are captured in real time by a camera and fed into a deep residual network for feature extraction. A deep residual network is an improved convolutional neural network that addresses the vanishing gradient problem in deep networks by introducing residual connections, enabling the extraction of finer and deeper image features. Based on these features, a pose estimation network is designed to predict the workpiece's pose. The workpiece image features extracted from the deep residual network are transformed through several fully connected layers, ultimately regressing the workpiece's pose vector. This pose vector typically includes six degrees of freedom: position and attitude (rotation angle). This regression task requires optimizing a loss function to minimize the error between the predicted pose vector and the workpiece's true pose. The output of the regression network is divided into two parts: a position vector and an attitude vector. The position vector describes the workpiece's coordinates in 3D space, while the attitude vector describes its rotation (expressed in Euler angles or quaternions). For various workpiece types, the network outputs pose estimates based on the input workpiece image features. A laser scanner or structured light camera acquires 3D point cloud data of the workpiece in real time, while simultaneously collecting 2D image data. The trained deep residual network continuously receives new image data and continuously updates the workpiece pose estimate in dynamic environments. By fusing 3D point cloud information with image features, the accuracy of pose estimation can be improved, especially in complex scenarios. By combining the real-time updated pose vector with 3D point cloud data, dynamic positioning and tracking can be implemented. Based on the current pose estimate, the workpiece's position and posture at the next moment can be predicted, and the robot's manipulation of the workpiece (such as grasping or placing) can be adjusted. In complex production environments, it can address issues such as workpiece movement, rotation, and occlusion, achieving high-precision positioning and tracking.

[0169] According to another embodiment of the present invention, Figure 2 As shown, a variable-scale moving target positioning and tracking device integrating multiple features is also provided, and the variable-scale moving target positioning and tracking device integrating multiple features includes:

[0170] Image processing module 1 is used to obtain an image of the workpiece to be marked, extract features from the workpiece image using a multi-scale feature fusion algorithm, obtain feature points of the workpiece image, and establish a feature point matching relationship between adjacent workpiece image frames;

[0171] Workpiece detection module 2 is used to perform preliminary detection on the contour of the workpiece image using Hough transform, determine the preliminary workpiece contour area, and classify the workpiece categories based on the preliminary workpiece contour area to obtain the final workpiece contour area;

[0172] Data analysis and regression model module 3, used to regress the feature points in the final workpiece contour area using the double point removal least squares method to determine the contour parameters;

[0173] The target positioning and tracking module 4 is used to obtain the three-dimensional point cloud and workpiece image features of the workpiece, estimate the workpiece pose vector using a deep convolutional neural network, and dynamically position and track the workpiece by combining the workpiece pose vector and the three-dimensional point cloud of the workpiece;

[0174] The image processing module 1 is connected to the workpiece detection module 2 and the data analysis and regression model module 3 , and the data analysis and regression model module 3 is connected to the target positioning and tracking module 4 .

[0175] In summary, with the help of the above technical solutions of the present invention, the present invention can capture the details and structures of the image at different scale levels through a multi-scale feature fusion algorithm. Such a method is extremely effective for detecting feature points that remain stable under scale transformation. This capability is particularly suitable for scenarios where the size, shape or orientation of the workpiece changes rapidly. The descriptor constructed by calculating the gradient and direction of each feature point around its surrounding pixels not only provides the location information of the point, but also accurately describes its local image features. The distance and similarity between the descriptors are used to match feature points, making the matching process more robust, especially when the image is slightly distorted or occluded due to the movement of the workpiece. It can then effectively connect the same workpiece in consecutive image frames, providing a stable input for dynamic tracking. The present invention enhances the accurate recognition of the workpiece contour by utilizing the Hough transform combined with multi-step edge detection and shape parameter extraction, especially when the image quality is interfered with by noise. Grayscale conversion and Gaussian filtering help reduce processing errors and ensure the accuracy of edge detection. The partitioning algorithm used in the classification process further optimizes the category recognition of workpieces, thereby ensuring that each type of workpiece can be correctly classified. The chaotic search technology is used to optimize and adjust the workpiece contour after the classification is determined. This not only improves the accuracy of tracking, but also increases the adaptability to dynamic changes. By integrating different solutions and adjusting algorithm parameters such as scaling factors and cross factors, it can flexibly respond to the diversity and complexity of the visual presentation of workpieces, thereby maintaining high performance in a constantly changing industrial environment. The present invention uses the least squares method for preliminary regression to provide a robust baseline, and by removing potential abnormal feature points pair by pair and recalculating the regression, it can significantly reduce the influence of noise and outliers, improve the accuracy and robustness of parameter estimation, and ensure the optimal performance of the model when facing variable and complex data through continuous analysis of difference values and selection of regression results with the smallest difference values. This method is particularly suitable for dynamically changing industrial environments, ensuring that the estimation of contour parameters can adapt to actual physical changes. By carefully managing each data point and iterative process, it can more efficiently process large amounts of data, reduce the waste of computing resources, and thus optimize operational efficiency and output quality. The present invention extracts high-quality features from workpiece images through a deep residual network and accurately calculates the pose vector through a pose estimation network, thereby achieving efficient dynamic tracking. In a dynamic environment, it can collect and process three-dimensional point clouds and image data in real time, and continuously update the pose estimate of the workpiece. This not only improves the system's adaptability to environmental changes, but also ensures that the continuity and accuracy of tracking can be maintained under various working conditions.

[0176] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A variable-scale moving target positioning and tracking method integrating multiple features, characterized by: include: S1. Acquire an image of the workpiece to be marked, extract features from the workpiece image using a multi-scale feature fusion algorithm, obtain feature points of the workpiece image, and establish a feature point matching relationship between adjacent workpiece image frames; S2. Performing preliminary detection on the contour of the workpiece image using Hough transform to determine a preliminary workpiece contour area, and classifying the workpiece categories based on the preliminary workpiece contour area to obtain a final workpiece contour area; including: S26. Based on the preliminary workpiece contour area, the workpiece category is divided using a division algorithm to obtain a final workpiece contour area; including: S261, setting the scaling factor and cross factor of the partitioning algorithm, and determining the maximum number of iterations and the number of chaotic searches; S262, classifying the workpiece into categories according to the characteristics of the workpiece image and initializing candidate solutions for the workpiece contour; S263, randomly selecting a workpiece contour candidate solution from the workpiece contour candidate solutions according to each workpiece category to generate a new variant workpiece contour candidate solution; S264, fusing different workpiece contour candidate solutions through cross factors to generate a new workpiece contour candidate solution; S265, comparing the newly generated candidate workpiece contour solution with the current optimal workpiece contour solution, and retaining the optimal workpiece contour solution; S266, repeating steps S263 to S265 until a predetermined number of iterations is reached; S267, performing a chaotic search around the optimal workpiece contour solution of each category, and selecting the workpiece contour with the highest fitness from the chaotic search as the final workpiece contour region; S3, using the double point removal least squares method to regress the feature points in the final workpiece contour area to determine the contour parameters; S4. Obtain the three-dimensional point cloud and workpiece image features of the workpiece, use a deep convolutional neural network to estimate the workpiece's pose vector, and combine the workpiece's pose vector with the workpiece's three-dimensional point cloud to perform dynamic positioning and tracking of the workpiece.

2. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 1 is characterized in that: The method of acquiring an image of the workpiece to be marked, extracting features from the workpiece image using a multi-scale feature fusion algorithm to obtain feature points of the workpiece image, and establishing a feature point matching relationship between adjacent workpiece image frames includes the following steps: S11, extracting feature points from the acquired workpiece image using a multi-scale feature fusion algorithm; S12, extracting stable feature points from the workpiece image, and calculating the descriptor of each feature point to characterize the image information around the feature point; S13, matching feature points in two adjacent workpiece images, and realizing feature point matching relationship by calculating the distance and similarity between the two descriptors.

3. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 2 is characterized in that: The method of extracting feature points from the acquired workpiece image using a multi-scale feature fusion algorithm comprises the following steps: S111, performing multi-scale transformation on the workpiece image by using a Gaussian difference pyramid, and generating workpiece images of different scale spaces; S112, detecting local extreme points as candidate feature points by comparing adjacent pixels in each scale-space workpiece image; S113, calculating the gradient and direction of the surrounding pixels for each candidate feature point, generating a gradient direction histogram, and determining the main direction of the feature point based on the peak value of the histogram; S114. Construct a local area with each feature point as the center, calculate the gradient direction and amplitude in the local area, and combine them into a descriptor vector as the descriptor of the feature point to perform feature matching.

4. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 1, characterized in that: The method of performing preliminary detection on the contour of the workpiece image by using Hough transform, determining a preliminary workpiece contour area, and classifying the workpiece categories based on the preliminary workpiece contour area to obtain the final workpiece contour area includes the following steps: S21, converting the workpiece color image into a grayscale image, and applying Gaussian filtering to remove image noise; S22, extracting edges in the grayscale image using an edge detection algorithm; S23, detecting the shape of the workpiece according to the need and setting the parameter space; S24, increasing the count of all parameters corresponding to each edge point on the grayscale image in the parameter space to form a cumulative array, and searching for a local maximum in the cumulative array; S25 , obtaining a peak value from the accumulated array as a parameter of the detected workpiece shape, and using the parameter of the workpiece shape to draw a contour on the original workpiece image as a preliminary workpiece contour area.

5. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 1 is characterized in that: The formula for generating a new candidate solution for the variant workpiece contour is: Ai(m+1)=Cr1(m)+G·(Cr2(m)-Cr3(m)); Where A i (m+1) represents the candidate solution of the i-th variant workpiece contour at the next iteration m+1; G represents the scaling factor; C r1 (m), C r2 (m) and C r2 (m) represents the candidate solutions of the workpiece contour randomly selected from the candidate solutions of the workpiece contour at the current iteration m; r1, r2 and r3 are random variables.

6. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 1, characterized in that: The method of regressing the feature points in the final workpiece contour area using the double-point removal least squares method to determine the contour parameters includes the following steps: S31, obtaining a feature point data set within the final workpiece contour area; S32, using the least squares method to perform preliminary linear regression analysis on all feature points to obtain initial contour parameter estimates; S33, removing two feature points in the feature point data set one by one, and re-performing least squares regression using the remaining feature points; S34, calculating the difference between the contour parameters obtained by least squares regression after removing two feature points each time and the initial contour parameter estimate, and recording the difference value; S35, repeating step S33 and step S34 until the preset number of iterations is reached; S36. Analyze the difference values of all records and select a set of regression results with the smallest difference values as the estimate of the final profile parameters.

7. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 6 is characterized in that: The step of analyzing the difference values of all records and selecting a set of regression results with the smallest difference values as the estimation of the final profile parameters includes the following steps: S361. Obtain all historical regression result datasets and initialize the regression result datasets and related constraints; S362, accepting new regression result data update instructions and selecting a data processing mode according to analysis requirements; S363. If the priority of minimizing the difference value is selected, the constraint boundary of the data set is calculated using a specific formula. If the priority of other is selected, the original data set boundary remains unchanged. S364. Calculate the deviation of each set of regression result data and the deviation of the entire regression result data set, and select data points that need to be adjusted first; S365. Perform regression analysis on the selected data points using an optimization algorithm, and record the difference value of each optimization; S366. Analyze the difference values of all records, compare the calculated difference values, and select a set of regression results with the smallest difference value as the final profile parameters.

8. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 7 is characterized in that: The method of performing regression analysis optimization on the selected data points using an optimization algorithm and recording the difference value of each optimization includes the following steps: S3651. Randomly initialize several sets of regression model parameters, each set of parameters representing a potential regression model solution; S3652. Calculate the fitness of each set of regression model parameters, and select the regression model with the smallest difference value as the core regression model; S3653. Adjust the non-core regression model parameter set, adopt different adjustment strategies for the core regression model parameter set, and determine the number of new parameters to be generated based on their respective difference values; S3654. Generate a new parameter set based on the current regression model parameters and perform boundary detection; S3655, applying Gaussian mutation to selected regression model parameters to generate additional parameter sets; S3656. From all current regression model parameters, select a set of parameters with the smallest difference value to continue iteration, and record the best difference value; S3657. Check whether the predetermined number of iterations has been reached. If so, stop the iteration and record the minimum difference value as the regression result.

9. The variable-scale moving target positioning and tracking method integrating multiple features according to claim 1, characterized in that: The method of acquiring the three-dimensional point cloud and image features of the workpiece, estimating the workpiece pose vector using a deep convolutional neural network, and dynamically positioning and tracking the workpiece by combining the workpiece pose vector and the three-dimensional point cloud of the workpiece includes the following steps: S41. Use deep residual network to extract features from workpiece images and design pose estimation network; S42, transforming the output of the feature extraction part through a fully connected layer to regress the pose vector; S43, dividing the output of the pose regression part into a position vector and a pose vector, outputting the estimated poses of the multiple classes of objects respectively; S44, collect 3D point cloud and image data of the workpiece in real time in a dynamic environment, and use the trained deep residual network to continuously update the workpiece pose estimation; S45. Combine the real-time updated pose vector and 3D point cloud to implement accurate dynamic positioning and tracking of workpieces.

10. A multi-feature integrated variable-scale moving target positioning and tracking device, used to implement the multi-feature integrated variable-scale moving target positioning and tracking method according to any one of claims 1 to 9, characterized in that: The multi-feature fusion variable-scale moving target positioning and tracking equipment includes: The image processing module is used to obtain the image of the workpiece to be marked, extract features from the workpiece image using a multi-scale feature fusion algorithm, obtain feature points of the workpiece image, and establish a feature point matching relationship between adjacent workpiece image frames; The workpiece detection module is used to perform preliminary detection on the contour of the workpiece image using Hough transform, determine the preliminary workpiece contour area, and classify the workpiece categories based on the preliminary workpiece contour area to obtain the final workpiece contour area; The data analysis and regression model module is used to regress the feature points in the final workpiece contour area using the double-point removal least squares method to determine the contour parameters; The target positioning and tracking module is used to obtain the workpiece's 3D point cloud and workpiece image features, estimate the workpiece's pose vector using a deep convolutional neural network, and dynamically position and track the workpiece by combining the workpiece's pose vector with the workpiece's 3D point cloud; Among them, the image processing module is connected through the workpiece detection module and the data analysis and regression model module, and the data analysis and regression model module is connected with the target positioning and tracking module.

Images